LGGTApr 29

Distributional Alignment Games for Answer-Level Fine-Tuning

arXiv:2604.2716695.41 citations
AI Analysis

This work provides a principled framework for optimizing language models based on answer correctness, unifying existing approaches and enabling efficient algorithms.

The authors propose a game-theoretic framework for answer-level fine-tuning of language models, transforming intractable marginalization into a tractable projection problem, and demonstrate significant complexity gains in mathematical reasoning tasks.

We focus on the problem of \emph{Answer-Level Fine-Tuning} (ALFT), where the goal is to optimize a language model based on the correctness or properties of its final answers, rather than the specific reasoning traces used to produce them. Directly optimizing answer-level objectives is computationally intractable due to the need to marginalize over the vast space of latent reasoning paths. To overcome this, we propose a general game-theoretical framework that lifts the problem to a \emph{Distributional Alignment Game}. We formulate ALFT as a two-player game between a Policy (the generator) and a Target (an auxiliary distribution). We prove that the Nash Equilibrium of this game corresponds exactly to the solution of the original answer-level optimization problem. This variational perspective transforms the intractable marginalization problem into a tractable projection problem. We demonstrate that this framework unifies recent approaches to diversity and self-improvement (coherence) and provide efficient algorithms compatible with Group Relative Policy Optimization (GRPO), such as Coherence-GRPO, yielding significant complexity gains in mathematical reasoning tasks.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes